Choose theme
AI Visuals
PersonaLive
PersonaLive animates a portrait using a reference image and a driving video, with offline and streaming inference paths.
GVCLab provides research code, pretrained weights and a Web UI. The repository now also links EditaLive as an updated project for full-body animation and editing; those are separate project capabilities. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
A portrait animation framework
A research framework for generating portrait animation from visual inputs.
Why it stands out
Streaming-style generation paths
Its streaming approach is aimed at longer video generation; the project reports a 12GB VRAM setup rather than requiring the entire video to be generated at once.
Availability
Repo, paper, weights, and scripts
Public code, paper, weights and documented inference scripts; running it requires the listed GPU environment.
Why it matters
What makes it useful
Use a driving video to supply movement for an animated portrait. The reference image supplies the portrait appearance, making the input workflow more specific than a text prompt alone.
What to know
Where it fits
The offline script takes a recorded video through inference; the online path supports streaming input. The Web UI is another entry point around the same research setup.
Notable points
What stands out
The repository includes TensorRT and xFormers acceleration notes. These are optional setup paths with their own environment requirements, rather than features that run automatically after downloading the weights. To animate a different portrait with the same driving-video input, the offline script accepts a replacement through --reference_image. This changes the reference appearance without asking for a new movement clip; the generated result still needs to be inspected.
Before using
What to review
Read the project’s research-use disclaimer and use portrait and driving-video material with appropriate permission. Its README says the project is for academic research only; also decide how a real person’s animated likeness will be used or presented.
Check the documented GPU, dependencies and weight-download paths before setting up inference. The project’s provided TensorRT engines were built on H100 and its notes recommend rebuilding for the local environment. Check the documented Blackwell/xFormers compatibility and inspect output quality when using acceleration.
Reader fit
Who may find it relevant
Researchers and builders experimenting with portrait animation or live-avatar workflows. It needs a configured inference environment, so it is less suited to someone seeking a ready-to-use video editor.
Editorial note
Why LifeHubber lists it
For research on an animated portrait driven by continuing incoming motion, the online worker prepares the portrait reference and reuses that state across successive frame batches. This gives researchers a streaming pipeline to study, with reference preparation separated from the ongoing frame processing; it does not promise a particular latency or likeness quality.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
More in Music / Image Gen Models
Keep browsing this category
Explore more media-generation model resources.
ACE-Step 1.5
ace-step/ACE-Step-1.5
A locally runnable music model with 2B and XL 4B variants for full-song generation, reference-guided creation, repainting, accompaniment, stem separation, and lightweight style training.
Qwen-Image-2.1
QwenLM/Qwen-Image-2.1
Qwen's open-weight image model for text-to-image generation, single- and multi-reference editing, marked-region changes, subject extraction, native transparent RGBA output, and 2K workflows.
Mage-Flow
microsoft/Mage-Flow
Mage-Flow documents text-to-image generation and instruction-based image editing in one Microsoft 4B family, with Base, RL-aligned, and four-step Turbo variants. Code is available; confirm checkpoint access before a local run, and check the hosted demo status.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
